fix: 第四轮调试 — 回复去重/消息时序/UI布局/自主思考深度优化 + 文档重整

后端修复:
- main.go: 恢复 /api/v1/chat 路由中丢失的 handleChat 调用 (空响应回归)
- orchestrator.go: splitChatByLines 改为双换行分割, 避免单换行误拆
- chat_handler.go: multi_message 增加 !hasReview 守卫, 消息延迟 200→800ms
- thinker.go: RecordUserMessage 追踪活跃会话ID, 推送主动消息到正确会话
- thinker.go: 增强思考提示词 — 禁止在用户休息/离开时发送主动消息

前端修复:
- useWebSocket.ts: stream_segments 不再创建消息气泡, 消除重复回复
- MessageBubble.tsx: 动作消息居左对齐无头像, 时间戳移至气泡外侧 hover 显示
- ChatInput.tsx: 昔涟输入提示移至输入框上方, 波点动画效果
- MessageList/TypingIndicator/ChatContainer: 清理冗余 isTyping 传递
- MemoryPanel.tsx: 新增记忆面板组件

文档重整:
- docs/debug/ → docs/debug_log/ 重命名统一
- 新增 debug_log/README.md 索引
- .gitignore: 新增 android/ 排除规则

Co-Authored-By: Claude Opus 4.7 <noreply@anthropic.com>

backend/ai-core/internal/orchestrator/orchestrator.go

@@ -131,16 +131,15 @@ func (o *Orchestrator) ProcessInput(
 			Nickname:      userName,
 		}
 
-		// 对于 simple greeting，跳过子会话分派，直接合成回复
-		var resultCh <-chan model.SubSessionResult
-		skipSubSessions := intent.Primary == "greeting" ||
-				(intent.Primary == "chat" && !intent.NeedsIoT && !intent.NeedsMemory)
-		if skipSubSessions {
-			log.Printf("[orchestrator] 快速通道: 简单消息(primary=%s)，跳过子会话分派", intent.Primary)
-			// 创建一个已关闭的空通道
-			emptyCh := make(chan model.SubSessionResult)
-			close(emptyCh)
-			resultCh = emptyCh
+		// 只有明确的关键词问候才跳过子会话分派，日常闲聊也需要检索记忆
+			// 因为 LLM 容易将日常闲聊误判为 needs_memory=false，导致回复缺乏上下文
+			var resultCh <-chan model.SubSessionResult
+			skipSubSessions := intent.Primary == "greeting" && !intent.NeedsMemory
+			if skipSubSessions {
+				log.Printf("[orchestrator] 快速通道: 简单问候(primary=%s)，跳过子会话分派", intent.Primary)
+				emptyCh := make(chan model.SubSessionResult)
+				close(emptyCh)
+				resultCh = emptyCh
 		} else {
 			resultCh = o.subManager.Dispatch(subCtx, intent, params.Message, createParams)
 		}
@@ -369,7 +368,7 @@ func parseReviewMessages(text string) []model.ReviewMessage {
 			if actionStart > 0 {
 				prefix := strings.TrimSpace(remaining[:actionStart])
 				if prefix != "" {
-					messages = append(messages, splitReviewLongMessage(model.ReviewMessageChat, prefix)...)
+					messages = append(messages, splitChatByLines(model.ReviewMessageChat, prefix)...)
 				}
 			}
 			// 括号内作为 action
@@ -385,7 +384,7 @@ func parseReviewMessages(text string) []model.ReviewMessage {
 			// 没有括号，剩余全部作为 chat
 			remaining = strings.TrimSpace(remaining)
 			if remaining != "" {
-				messages = append(messages, splitReviewLongMessage(model.ReviewMessageChat, remaining)...)
+				messages = append(messages, splitChatByLines(model.ReviewMessageChat, remaining)...)
 			}
 			break
 		}
@@ -409,6 +408,26 @@ func splitReviewLongMessage(msgType model.ReviewMessageType, text string) []mode
 	if len(runes) <= maxLen {
 		return []model.ReviewMessage{{Type: msgType, Content: text}}
 	}
+	// ... split by sentence boundaries for long messages
+	return splitLongText(msgType, runes, maxLen)
+}
+
+// splitChatByLines 将聊天文本按双换行（段落分隔）拆分为多条消息，每条再检查是否需要按长度拆分
+func splitChatByLines(msgType model.ReviewMessageType, text string) []model.ReviewMessage {
+	lines := strings.Split(text, "\n\n")
+	var msgs []model.ReviewMessage
+	for _, line := range lines {
+		line = strings.TrimSpace(line)
+		if line == "" {
+			continue
+		}
+		msgs = append(msgs, splitReviewLongMessage(msgType, line)...)
+	}
+	return msgs
+}
+
+// splitLongText 将文本按句子边界分割
+func splitLongText(msgType model.ReviewMessageType, runes []rune, maxLen int) []model.ReviewMessage {
 
 	var messages []model.ReviewMessage
 	start := 0
@@ -459,7 +478,7 @@ func splitReviewLongMessage(msgType model.ReviewMessageType, text string) []mode
 	if len(messages) == 0 {
 		messages = append(messages, model.ReviewMessage{
 			Type:    msgType,
-			Content: text,
+			Content: string(runes),
 		})
 	}